Pricing
One subscription. Every model.
Both plans include every model, unlimited context, and unlimited concurrency. The only difference is the size of your compute budget: start with Spark, and step up to Ember when you need more usage.
Spark
The plan for most people.
A weekly compute allowance for everyday inference.
1x weekly allowance
A budget that goes a very long way on fast, efficient models. Every other model in the catalog is included too.
Cancel anytime · billed monthly
Ember
The same plan, twice the budget.
Twice Spark's weekly compute allowance for heavier workloads.
2x weekly allowance
Room to lean on the larger models all day, every day. Every other model in the catalog is included too.
Cancel anytime · billed monthly
Spending a lot of time in premium models like Claude? One-time top-up credits stack on either plan, start at $5, and never expire.
Browse credit packsHow your balance works
One weekly allowance, no surprises.
No token math, no surprise invoices. Everything you run draws from one weekly allowance that resets every Friday. Out of room before the reset? Top-up credits stack on top anytime, which comes in handy if you lean on premium models like Claude.
Weekly allowance
resets every FridayYour plan includes a weekly usage allowance. The allowance resets every Friday at a predictable time (and refills when your plan renews). Unused allowance does not roll over, and anything spent beyond it is deducted from the next reset.
Top-up credits
never expireOne-time credit packs that stack on top of your plan. Requests draw from them only after your weekly allowance is spent, and they never expire: whatever you buy stays yours until you use it.
Usage
What affects request usage?
Every request uses compute from your plan balance, and some requests use more than others. This is expected: bigger or more complex requests cost more to run.
- The model you choose. Heavier models cost more per request.
- How much conversation context is included. More context means more compute.
- Long conversations and unlimited-context use. They keep growing as you chat.
- Output length. Longer replies use more balance.
- Attachments and uploaded files. Images and documents add to each request.
- Tool and search usage. Lookups and tool calls add a little extra.
- Running multiple requests at once. Concurrent generations spend balance in parallel.
You don't need to calculate tokens manually. KI handles the metering and shows your remaining balance clearly, with no surprise invoices. And if a heavy week on models like Claude runs you low, top-up credits fill the gap without touching your plan.